Open adler-j opened 8 years ago
Looks very nice indeed.
There is a branch for this out now. Compilation times are down drastically and we no longer need a boost dependency, instead we have PyBind11 as a submodule, but there are some issues. It does seem PyBind11 is actively maintained so they could be fixed soon.
EDIT: 5 tests still fail, one serious (index error not raised), and a few less serious (getitem returns a long
in python 2).
Sounds great so far. I'm sure the errors can be fixed upstream.
Now 100% working, biggest (and really only) issue is that C++11 is not fully supported with VS2013, and CUDA does not work with VS2015. Currently, it works, so we can freeze at this commit of PyBind11 if we want, but if the author does some change that breaks VS2013, we cannot use that until CUDA catches up.
Ah these windows issues... Of course, we can freeze at the current state. When there is an upstream change, it's easy to just test and see if our systems cooperate with it.
I don't know if this conversation contains the latest issue which made us stall this, but now that CUDA 8 is out and VS15 support is there, perhaps it's time to reconsider?
It certainly is. I'll look more at it in a while.
As for now, CUDA 8 hasn't hit the mainstream Linux distro repos yet, so there's no need for a rush.
PyBind11 is an alternative to boost python that does not require a compiled part (and the rest of boost), this would simplify compilation and distribution by PyPI.